high-throughput phenotyping of maize leaf physiological ... · high-throughput, noninvasive ﬁeld...

High-Throughput Phenotyping of Maize LeafPhysiological and Biochemical Traits UsingHyperspectral Reflectance1[OPEN]

Craig R. Yendrek, Tiago Tomaz, Christopher M. Montes, Youyuan Cao, Alison M. Morse, Patrick J. Brown,Lauren M. McIntyre, Andrew D. B. Leakey, and Elizabeth A. Ainsworth*

Carl R. Woese Institute for Genomic Biology (C.R.Y., T.T., C.M.M., Y.C., P.J.B., A.D.B.L., E.A.A.), Departmentof Plant Biology (C.M.M., A.D.B.L., E.A.A.), and Department of Crop Sciences (P.J.B.), University of Illinois atUrbana Champaign, Urbana, Illinois 61801; Plant Protection Department, Fujian Agriculture and ForestryUniversity, Fuzhou 350002, China (Y.C.); Department of Molecular Genetics and Microbiology and GeneticsInstitute, University of Florida, Gainesville, Florida 32610 (A.M.M., L.M.M.); and Global Change andPhotosynthesis Research Unit, United States Department of Agriculture-Agricultural Research Service, Urbana,Illinois 61801 (E.A.A.)

ORCID ID: 0000-0002-3199-8999 (E.A.A.).

High-throughput, noninvasive field phenotyping has revealed genetic variation in crop morphological, developmental, andagronomic traits, but rapid measurements of the underlying physiological and biochemical traits are needed to fully understandgenetic variation in plant-environment interactions. This study tested the application of leaf hyperspectral reflectance (l = 500–2,400 nm) as a high-throughput phenotyping approach for rapid and accurate assessment of leaf photosynthetic and biochemicaltraits in maize (Zea mays). Leaf traits were measured with standard wet-laboratory and gas-exchange approaches alongsidemeasurements of leaf reflectance. Partial least-squares regression was used to develop a measure of leaf chlorophyll content,nitrogen content, sucrose content, specific leaf area, maximum rate of phosphoenolpyruvate carboxylation, [CO2]-saturated rateof photosynthesis, and leaf oxygen radical absorbance capacity from leaf reflectance spectra. Partial least-squares regressionmodels accurately predicted five out of seven traits and were more accurate than previously used simple spectral indices for leafchlorophyll, nitrogen content, and specific leaf area. Correlations among leaf traits and statistical inferences about differencesamong genotypes and treatments were similar for measured and modeled data. The hyperspectral reflectance approach tophenotyping was dramatically faster than traditional measurements, enabling over 1,000 rows to be phenotyped during middayhours over just 2 to 4 d, and offers a nondestructive method to accurately assess physiological and biochemical trait responses toenvironmental stress.

Meeting future global food demand is projected torequire a doubling of agricultural output by 2050 (Tilmanet al., 2011). Climate changewill exacerbate this challengeby intensifying the exposure offield crops to abiotic stressconditions, including rising temperature, increaseddrought, flooding, and air pollution (Christensen et al.,2007). In order to more rapidly adapt crops to the future

growing environment, diverse germplasm must bemeaningfully assessed in field conditions. However,our capacity to phenotype thousands of crop genotypesin a field environment is constrained by available high-throughput phenotyping platforms (Furbank andTester, 2011; Araus and Cairns, 2014), and our abilityto adapt crops to global climate change requires fa-cilities that accommodate hundreds of genotypes ina realistic environment (Ainsworth et al., 2008).

This study provides proof of concept for one approachfor high-throughput phenotyping of leaf physiologicalresponses to global change using hyperspectral reflectancespectra of diversemaize (Zeamays) inbred andhybrid linesgrown at ambient and elevated ozone concentrations([O3]) in the field. Hyperspectral sensors capture electro-magnetic radiation reflected from vegetation in the visible(400–700 nm), near-infrared (700–1,300 nm), and short-wavelength infrared (1,400–3,000 nm) regions, whichcontain information about leaf physiological status, in-cluding pigments, structural constituents of biomass, andwater content (Curran, 1989; Martin and Aber, 1997;Penuelas and Filella, 1998). Vegetation spectra can becaptured in a few seconds, and universal indices have

1 This work was supported by the National Science Foundation(grant no. PGR–1238030) and by a Fulbright Western Australia Schol-arship (to T.T.).

* Address correspondence to [email protected] author responsible for distribution of materials integral to the

findings presented in this article in accordance with the policy de-scribed in the Instructions for Authors (www.plantphysiol.org) is:Elizabeth A. Ainsworth ([email protected]).

C.R.Y., E.A.A., P.J.B., L.M.M., and A.D.B.L. conceived of and de-signed the original research plans; C.R.Y., T.T., and C.M.M. per-formed most of the experiments; Y.C. provided technical assistanceto C.R.Y.; C.R.Y., A.M.M., L.M.M., and E.A.A. analyzed the data;C.R.Y. and E.A.A. wrote the article with contributions from all ofthe authors.

[OPEN] Articles can be viewed without a subscription.www.plantphysiol.org/cgi/doi/10.1104/pp.16.01447

614 Plant Physiology�, January 2017, Vol. 173, pp. 614–626, www.plantphysiol.org � 2017 American Society of Plant Biologists. All Rights Reserved. www.plantphysiol.orgon May 26, 2020 - Published by Downloaded from

Copyright © 2017 American Society of Plant Biologists. All rights reserved.

http://orcid.org/0000-0002-3199-8999

mailto:[email protected]

http://www.plantphysiol.org

mailto:[email protected]

http://www.plantphysiol.org/cgi/doi/10.1104/pp.16.01447


been proposed to estimate leaf foliar traits such as chlo-rophyll content, nitrogen (N) content, and photosyntheticradiation use efficiency from reflectance data (Gamonet al., 1997; le Maire et al., 2004; Herrmann et al., 2010).Predictive models using partial-least squares regression(PLSR) approaches to scale spectral data also have beenapplied to estimate crop yield (Weber et al., 2012) andto estimate leaf features, including isotopic ratios(Richardson and Reeves, 2005), the ratio of leaf mass toleaf area (Asner and Martin, 2008; Asner et al., 2011),carbohydrate content of plant tissues (Dreccer et al.,2014; Asner and Martin, 2015), and the maximumphotosynthetic capacity of C3 plants (Serbin et al., 2012;Ainsworth et al., 2014). Variation in foliar reflectance atdifferent wavelengths in the spectrum is specific tovariation in different chemical and structural compo-nents of leaves (Curran, 1989; Serbin et al., 2014).Therefore, analysis of foliar reflectance spectra has thepotential to very rapidly assess multiple physiologicaland biochemical traits from a single measurement.The use of hyperspectral reflectance as a high-

throughput phenotyping tool in agricultural researchis promising (Weber et al., 2012; Araus and Cairns,2014), but questions remain regarding limitations of thetechnique. Reflectance spectroscopy has been used tosuccessfully distinguish broad-scale spatial and tem-poral variation in leaf foliar traits and photosyntheticmetabolism across species in diverse agroecosystems(Serbin et al., 2015), tropical ecosystems (Asner andMartin, 2008; Asner et al., 2011), and northern temper-ate and boreal forests (Serbin et al., 2014). In this study,we test the utility of leaf hyperspectral reflectance as ahigh-throughput method for phenotyping diversegermplasm subject to abiotic stress in maize. Maize isone of the world’s most important crops and also rep-resents a model for species with C4 photosynthesis. Therelationship of hyperspectral reflectance with C4 leaftraits has the potential to be distinct from that of C3species due to the significantly different leaf anatomy,biochemistry, and physiology of the two photosyn-thetic types.We grew diverse maize genotypes in the field under

ambient and elevated [O3] using Free Air ConcentrationEnrichment (FACE). O3 is an air pollutant that entersleaves through the stomata, where it rapidly formsother reactive oxygen species that cause damage to bi-ochemical constituents and physiological processes,ultimately leading to reductions in leaf longevity andyield (Fiscus et al., 2005; Ainsworth et al., 2012). Pastefforts to detect the genetic variation in O3 toleranceusing mapping populations have scored visible leafdamage or plant biomass (Kim et al., 2004; Frei et al.,2008; Brosché et al., 2010; Street et al., 2011; Ueda et al.,2015) but have not quantified the many other physio-logical changes that ultimately determine the impact ofO3 on crop yield. These include reduced photosyntheticcapacity, reduced leaf chlorophyll and N content, accel-erated senescence, increased antioxidant capacity, alteredcarbohydrate and metabolite content, decreased specificleaf area (SLA), and increased rates of respiration (Fiscus

et al., 2005; Leitao et al., 2007; Dizengremel et al., 2008;Betzelberger et al., 2012; Gillespie et al., 2012). The lo-gistics and cost of employing traditional phenotypingmethods to quantify these physiological and biochemicalmarkers of stress are prohibitive at the scale needed toscreen large populations, so hyperspectral reflectanceoffers a promising nondestructive, rapid, and high-throughput alternative. However, this alternative ap-proach has yet to be validated.

In order to test the throughput of this approach in thefield, we collected reflectance spectra from diverse inbredand hybrid maize lines grown in ambient and elevated[O3] in replicated plots across a 16-ha field over threeconsecutive growing seasons as well as greenhouse-grown maize (inbred B73) given optimal and limiting Nsupply. This range of conditions and material was devel-oped to ensure that awide range of valuesweremeasured.These same leaves were sampled using gold standardtechniques of gas-exchange and wet-laboratory biochem-ical approaches to quantify traits relevant to leaf stressphysiology, including chlorophyll content, N content,SLA, maximum rate of phosphoenolpyruvate carboxyla-tion (Vp,max), [CO2]-saturated rate of photosynthesis (Vmax),leaf oxygen radical absorbance capacity (ORAC), and Succontent. The two data sets were used successfully to buildpredictive PLSR models for five of seven traits and todemonstrate that similar conclusions about genetic andtreatment variation in leaf traits were drawn from bothmeasured and modeled data.

RESULTS

Predicting Physiological Traits from Reflectance Spectra

PLSRmodels using the leave-one-out cross-validationapproach predicted leaf biochemical and physiologicaltraits with varying degrees of accuracy. The predictiveability was strongest for chlorophyll (r2 = 0.81–0.85; Fig.1,A andB) andN content (r2 = 0.92–0.96; Fig. 1, C andD),good for SLA (r2 = 0.68–0.78; Fig. 1, E and F), Vmax (r

2 =0.56–0.65; Fig. 1, G andH), and Suc content (r2 = 0.62; Fig.2A), and nonsignificant for Vp,max (Fig. 1, I and J) andORAC (Fig. 2B). In general, models created with spectracollected from leaves exposed to multiple growth envi-ronments (i.e. greenhouse and field, low and highN, andambient and elevated [O3]) captured a greater propor-tion of trait variation than models built from just field-grown plants. For example, a stronger correlation be-tween measured and model-predicted chlorophyllvalues was observed with the combined field andgreenhouse model (r2 = 0.85; Fig. 1B) compared with thefield-only model (r2 = 0.81; Fig. 1A). Similarly, the cor-relations for N content and Vmax were stronger usingboth greenhouse and field data. In contrast, inclusion ofdata from greenhouse-grown maize plants did not im-prove the SLA model (Fig. 1F). No greenhouse sampleswere collected for ORAC and Suc quantification; there-fore, models consisted only of data from the field (Fig. 2).PLSR model coefficients for each trait are provided inSupplemental Table S1.

Plant Physiol. Vol. 173, 2017 615

High-Throughput Phenotyping of Maize Leaf Traits

www.plantphysiol.orgon May 26, 2020 - Published by Downloaded from Copyright © 2017 American Society of Plant Biologists. All rights reserved.

http://www.plantphysiol.org/cgi/content/full/pp.16.01447/DC1


As an additional test of PLSR model stability, 20% ofthemeasured values used for training the original PLSRmodels were randomly selected and removed (as aholdout data set). A second set of PLSR models wasbuilt using the remaining 80% of the original trainingdata (80% model). The parameter estimates generatedby the 80% model were very similar to the parameterestimates from the original, full model (SupplementalTable S2). In addition, the 80% model accurately esti-mated the measured values of the holdout data set thatwere removed prior to model building (SupplementalTable S2).

In order to test the similarity in relationships amongtraits for measured and PLSR-modeled data, pairwisecorrelations among traits were examined (Fig. 3). Inboth measured and PLSR-modeled data sets, the cor-relations were consistent, demonstrating that the rela-tionships among the underlying measured traits wereaccurately reflected by the PLSRmodels. N content andVmax were positively correlated, and there were signif-icant correlations between chlorophyll content and Ncontent and Vmax (Fig. 3). N content was negativelycorrelated to Suc content, and chlorophyll content wasnegatively correlated to SLA (Fig. 3).

Figure 1. Measured versus PLSR-modeled values of leaf chlorophyll content (A and B), N content (C andD), SLA (E and F),Vmax (Gand H), and Vp,max (I and J). PLSR models were built using a training population consisting of genotypically diverse field-grownplants exposed to ambient and elevated [O3] (A, C, E, G, I) or a combination of field-grown plants and greenhouse plants(genotype B73) grown in ample or limiting N (B, D, F, H, J). In each graph, the cross-validation r2, root mean square error (RMSE),and bias of the model are shown along with the size of the validation population and the number of model components (comp)used in each PLSR model. The gray dashed line shows the 1:1 line.

616 Plant Physiol. Vol. 173, 2017

Yendrek et al.







PLSR models were able to distinguish the same effectsas measured traits. Using genotypes B73, Mo17, and theB73 3 Mo17 hybrid, ANOVA models estimating the ef-fects of genotype, O3 treatment, and the interaction hadconsistent inferences for three of four successful PLSRmodels (chlorophyll, N, and Vmax; Supplemental TableS3; Supplemental Fig. S1). For SLA, the P value for the O3effect onmeasured datawas 0.012 and the P value for thePLSR-modeled data was 0.095 (Supplemental Table S3).To assess the relative contributions of different

wavelengths across themeasured spectra as drivers of agiven PLSR model, variable importance in projection(VIP) scores for each trait model were compared. The

chlorophyll model was very sensitive to variation inreflection of visible wavelengths, with major peaks at554 and 719 nm andminor contributions from the short-wavelength infrared region (Fig. 4A). The N contentmodel was built with wavelengths between 1,500 and2,400 nm based on the dominant absorption features ofN (Curran, 1989; Serbin et al., 2014) andwas sensitive tovariation in reflectance throughout that region of thespectrum. The SLA (Fig. 4B) and Suc content (Fig. 4C)models both had influential peaks in the visible wave-lengths that overlappedwith the chlorophyll model butalso showed trait-specific contributions throughoutthe near- and short-wavelength infrared regions. Thechlorophyll and Vmax models were very similar in thevisible portion of the spectrum, with moderate differ-ences between 800 and 1,750 nm (Fig. 4D).

As an alternative to PLSRmodel building, chlorophyll,N, and SLA also were estimated using previously pub-lished simple indices and correlatedwith directmeasuresof traits (Table I). Several indices, including mNDVI,mSRCHL, SIPI, and SR2,were unable to predict chlorophyllcontent accurately in maize leaves (r2 = 0.064, 0.006,0.067, and 0.172). Estimated chlorophyll values fromother indices were correlated with measured values(Table I), but even the best of these indices, based on thecoefficient of determination (Datt4; r2 = 0.74), was out-performed by the PLSR model (r2 = 0.85; Fig. 1B). Traitvalues estimated using previously published indices forN content and SLA showed poor correlations withmeasured values and were outperformed by PLSRanalysis of reflectance spectra (Table I; Fig. 1).

High-Throughput Screening of Diverse Inbred and HybridMaize for O3 Response

Once it was determined that PLSR models couldreasonably predict foliar biochemical and physiologicaltraits, we tested the approach for high-throughputphenotyping potential by screening diverse inbredand hybrid maize lines grown in the field at ambi-ent andelevated [O3] in 2013, 2014, and2015 (SupplementalTable S4). Using two spectroradiometers, we were able tocollect leaf reflectance spectra from maize plants grown in1,024 rows spread across a 16-ha field in 2 to 4 d whilelimiting the period of measurement to the middle 4 h ofeach day (approximately 11 AM to 3 PM). By contrast, de-structive sampling from each of the rows of the experimentrequired 12 people to sample for 4 to 5 h over 2 d, makingsample collection approximately 6 times more efficientfor spectral traits. The time savings of spectral reflec-tance over wet chemistry were even more dramaticfor postprocessing of samples and data. Analysis ofspectral data took approximately 10 d, while processingand analysis of samples for wet chemistry and leaf gasexchange took between 30 and 90 d depending onthe trait.

Diverse inbred and hybrid maize genotypes showeda 1.3- to 2.2-fold range of values for chlorophyll content,N content, SLA, Suc content, and Vmax, as estimatedby the PLSR models (Fig. 5). For example, in 2014,

Figure 2. Measured PLSR-modeled values of leaf Suc content (A) andleaf total ORAC (B). PLSR models were built using a training populationconsisting of diverse field-grownmaize inbred and hybrid lines exposedto ambient and elevated [O3]. In each graph, the cross-validation r2,RMSE, and bias of the model are shown along with the size of thevalidation population and number of model components (comp) usedin each PLSR model. The gray dashed line shows the 1:1 line.











chlorophyll content ranged from 15.2 to 31.2 mg cm22

(Fig. 5A), N content from 3.2% to 4.3% (Fig. 5B), SLAfrom 20.6 to 29.6 mm2 mg21 (Fig. 5C), Suc content from3.8 to 5.7 mmol cm22 (Fig. 5D), and Vmax from 18.8 to41.1 mmol m22 s21 (Fig. 5E) in the 52 inbred genotypesgrown at ambient [O3]. In general, the range of traitvalues for hybrids grown in ambient [O3] was narrowerthan for inbreds grown at ambient [O3] (Fig. 5), as maybe expected because all of the hybrids shared a commonparent in B73 (Supplemental Table S4). Hybrid geno-types also had greater average chlorophyll, N, andVmaxcompared with inbred genotypes and lower averageSuc content (Fig. 5D). Inbred and hybrid genotypesgrown at elevated [O3] showed significant reductions inchlorophyll content in all years (Fig. 5A; Table II), and

while there was some interannual variation, elevated[O3] also decreased N content and Vmax (Fig. 5, B and E;Table II) and increased SLA (Fig. 5C; Table II).

DISCUSSION

This study demonstrates the ability to rapidly andcost-effectively screen for intraspecific variation inmaize leaf biochemical and physiological traits usingPLSR models applied to leaf reflectance spectra. Leafchlorophyll content, N content, SLA, Suc content, andVmax could be predicted from reflectance spectra ofmaize (Figs. 1 and 2). This capability to rapidly andsimultaneously capture five of the most commonlyassessed and important leaf traits to plant carbon andN

Figure 3. Pairwise correlations of measured (black symbols) and modeled (gray symbols) leaf traits. Modeled traits are partialleast-squares predictions from leaf reflectance spectra collected from diverse maize inbred and hybrid lines. Chl, Chlorophyll.


Yendrek et al.




relations in a single spectrum represents a significantadvance with a wide range of potential applications inmaize genetics and breeding. Previous attempts touse remote sensing to investigate O3 damage in maizemeasured visual damage to leaves and canopies(Runeckles and Resh, 1975; Kraft et al., 1996; Rudorffet al., 1996; for review, see Meroni et al., 2009), mostoften in terms of the normalized difference vegetationindex,which is based on differences in surface reflectancein the near-infrared and visible ranges of the spectrum.This was also limited to canopy-level remote sensing of asingle maize genotype. This study demonstrated successin the early detection and evaluation of leaf-level bio-chemical and physiological responses when visualsymptoms of O3 stress were minimal. The proof ofconcept for the use of spectral reflectance to detectphenotypic variation within a C4 species is a significantadvance over previous uses of similar techniques thathave distinguished larger species-level differences innatural or agricultural ecosystems (Asner and Martin,2008; Asner et al., 2011; Serbin et al., 2012, 2014, 2015).

High-Throughput Phenotyping withHyperspectral Reflectance

High-throughput field phenotyping efforts havesuccessfully investigated morphological, developmen-tal, and agronomic traits in large numbers of crop

genotypes. However, rapid measurements of physio-logical traits are needed to fully understand plant-environment interactions, including the plant stressresponse (Großkinsky et al., 2015). For field phenotyp-ing to be maximally useful, approaches should be highthroughput, precise, and multidimensional, providingaccurate information about relevant traits (Dhondtet al., 2013). The hyperspectral reflectance and PLSRmodeling approach taken in this study meets thosegoals. The collection of leaf reflectance spectra takes lessthan 1 min per leaf and is nondestructive, so it is pos-sible to monitor a leaf over its lifetime. To estimate Vmaxfrom gas exchange, it could take at least 20 to 30 min toassess the response of net assimilation (A) to variationin intercellular [CO2] (Ci) curve on a single leaf. Evenmeasurements of maximum light and CO2-saturatedphotosynthesis could take 5 to 10 min per leaf in thefield, where the leaf has to acclimate to the cuvetteconditions, which may differ from atmospheric condi-tions. Generating parameter estimates by fitting aphotosynthesis model to the measured gas-exchangedata would take approximately the same amount oftime as fitting the PLSR model. Therefore, we estimatethat the reflectance approach for estimating Vmax is atleast 10 times faster in terms of data collection and canbe completed during a shorter period of time during theday to avoid circadian or diurnal effects on leaf traitsthat might confoundmeasurements taken over a longer

Figure 4. VIPacross the spectral region for eachPLSR model. VIP scores for the chlorophyllmodel are comparedwith models for N content(%; A), SLA (B), Suc (C), and Vmax (D). A VIPscore greater than 0.8 is considered to influencethe trait estimate (Wold, 1994). The orangeshaded area in A indicates that the model for Ncontent was built using a truncated spectralregion (1,500–2,400 nm). All other modelsutilized the spectrum from 500 to 2,400 nm.





period of time (Dodd et al., 2005). For other traits, likeSuc content or N content, gold standard measurementsrequire a destructive leaf sample for biochemical anal-ysis, which is lower throughput and does not allowrepeated monitoring over time. Estimation of numer-ous traits from a single spectrum provides multidi-mensionality, and the speed with which spectra can becollectedmeans that, at least temporally,measurementscan be taken with high resolution. In this study, wederived leaf-level biochemical and structural traits fromleaf reflectance spectra collected with a leaf clip thatprovided a uniform light environment, thus minimiz-ing some of the environmental and spatial variationthat canopy-level reflectance measurements inevitablyhave. The leaf clip facilitates the investigation of tissue-level phenomena and the effects of leaf age, but there isa tradeoff in that further studywould be needed to scalethese leaf-level parameters to a canopy.

Leaf-Level Predictive Models

Leaf-level hyperspectral data have been used previ-ously to estimate photosynthetic parameters in C3 species(Doughty et al., 2011; Serbin et al., 2012) but not in C4species, which have a carbon-concentrating mechanismand different biochemical limitations to photosyntheticrate than C3 species. For example, while light-saturatedphotosynthetic rate is limited by carboxylation capacity

at low intercellular [CO2] in C3 plants (Farquhar et al.,1980), it is instead limited by phosphoenolpyruvate car-boxylase activity at low intercellular [CO2] in C4 plants(von Caemmerer, 2000). C3 and C4 species also investdifferentially in photosynthetic proteins and pigmentsand have leaves with distinct morphological and struc-tural characteristics. Based on these differences, a numberof studies have used remote sensing approaches to dis-tinguish C3 and C4 species experimentally or on thelandscape (Irisarri et al., 2009; Liu and Cheng, 2011;Adjorlolo et al., 2012). Perhaps it is not surprising that anew PLSR model was needed to estimate parametersrelated to C4 photosynthesis in maize, especially giventhat predictions of photosynthetic capacity are not directexpressions of the reflectance spectra but, instead, are dueto a combination of anatomical (e.g. SLA) and chemical(e.g. N or chlorophyll) factors that are expressed moredirectly in reflectance spectra (Jacquemoud and Baret,1990). In our study, we could not use reflectance spectrato accurately predict Vp,max in maize (Fig. 1, I and J) butcould accurately predict Vmax (Fig. 1, G and H). It willbe interesting to test if this is also the case for other C4species.

Similar to previous PLSR models for C3 photosyn-thetic parameters (Serbin et al., 2012), the C4 PLSRmodel for Vmax used wavebands in the visible region ofthe spectrum where chlorophyll and other pigmentshave strong features but also used wavebands in the

Table I. Estimates of chlorophyll content, N content, and SLA calculated using published spectral indices and compared with measured values

Each index was applied to the same spectra used for the PLSR model training population. For each index, the predictive formula is shown alongwith the correlation coefficient, RMSE, and reference.

Trait Index Formulation Adjusted r2 RMSE Reference

Chlorophyll Datt4 R672/(R550 3 R708) 0.740 3.23 Datt (1998)Vogelmann2 (R734 2 R747) 2 (R715 + R726) 0.736 3.26 Vogelmann et al. (1993)Maccioni (R780 2 R710)/(R780 2 R680) 0.714 3.39 Maccioni et al. (2001)Double Difference (R749 2 R720) 2 (R701 2 R672) 0.702 3.46 le Maire et al. (2004)Vogelmann1 R740/R720 0.701 3.47 Vogelmann et al. (1993)mSR705 (R750 2 R445)/(R705 2 R445) 0.680 3.58 Sims and Gamon (2002)mNDVI705 (R750 2 R705)/(R750 + R705 2 2R445) 0.653 3.73 Sims and Gamon (2002)SR3 R750/R550 0.580 4.11 Gitelson and Merzlyak

(1997)SR4 R700/R670 0.498 4.49 McMurtey et al. (1994)SR1 R750/R700 0.457 4.67 Gitelson and Merzlyak

(1997)Gitelson 1/R700 0.387 4.96 Gitelson et al. (1999)SR2 R752/R690 0.172 5.77 Gitelson and Merzlyak

(1997)SIPI (R800 2 R445)/(R800 2 R680) 0.067 6.12 Penuelas et al. (1995)mNDVI (R800 2 R680)/(R800 + R680 2 2R445) 0.064 6.13 Sims and Gamon (2002)mSRCHL (R800 2 R445)/(R680 2 R445) 0.006 6.36 Sims and Gamon (2002)

N NRI1510 (R1510 2 R660)/(R1510 + R660) 0.450 0.90 Herrmann et al. (2010)MCARI1510 [(R700 2 R1510) 2 0.2(R700 2 R550)] 3 (R700/R1510) 0.239 1.05 Herrmann et al. (2010)NDNI [log10(1/R1510) 2 log10(1/R1680)]/[log10(1/R1510) + log10(1/R1680)] 0.198 1.08 Serrano et al. (2002)

SLA mND (R2285 2 R1335)/[R2285 + R1335 2 (2 3 R2400)] 0.177 2.43 le Maire et al. (2008)NDMI (R1649 2 R1722)/(R1649 + R1722) 0.133 2.49 Cheng et al. (2014)NDLMA (R1368 2 R1722)/(R1368 + R1722) 0.122 2.51 Cheng et al. (2014)mSRSLA (R2265 2 R2400)/(R1620 2 R2400) 0.073 2.58 le Maire et al. (2008)SR R2295/R1500 0.070 2.58 le Maire et al. (2008)ND (R1710 2 R1340)/(R1710 + R1340) 0.061 2.60 le Maire et al. (2008)


Yendrek et al.



near-infrared and short-wavelength infrared regions(Fig. 4), indicating that Vmax is not simply a function ofchlorophyll content. We also observed, along withprevious studies (Hansen and Schjoerring, 2003;Doughty et al., 2011; Serbin et al., 2012), that both themeasured and PLSR-modeled leaf traits were signifi-cantly correlated (Fig. 3). Relationships among leafstructural and biochemical traits also converge across avery diverse range of species, and photosynthesis ispositively correlated to leaf N content and SLA (Reichet al., 1997; Wright et al., 2004). Global trait databasesalso have reported correlations between leaf N contentand SLA (Reich et al., 1997); however, this correlationwas not significant within diverse maize lines (Fig. 3).This may be because within-species variation is moreconstrained than across-species variation in leaf traits.Most importantly, the same conclusions about correla-tion among traits and variation in leaf traits due togenotype or treatment were drawn from tests of

measured and modeled data (Fig. 3; Supplemental Fig.S1; Supplemental Table S3).

PLSR models developed for maize also improvedpredictions of chlorophyll content, N content, and SLAcompared with previously published simple indices(Table I). The performance improvement of the PLSRmodel compared with simple indices has been reportedpreviously (Hansen and Schjoerring, 2003; Atzbergeret al., 2010) and has several contributing sources. First,PLSR takes advantage of the full spectral information,most of which is excluded by simple indices, yet clearlycontributes to improved trait prediction based on un-derlying biophysical or biochemical attributes (Hansenand Schjoerring, 2003). PLSR also is relatively insensitiveto sensor noise (Atzberger et al., 2010). The other ad-vantage is in building models specifically to the speciesunder study, in this case maize, while the other simpleindices were built using other species or diverse species(Vogelmann et al., 1993; Datt, 1998; Maccioni et al., 2001;

Figure 5. Box plots showing the distribution of values estimated from PLSR models for diverse maize genotypes, including foliarchlorophyll content (A), N content (B), SLA (C), Suc content (D), and Vmax (E). Maize inbred (IN) and hybrid (HY) genotypes weregrown at ambient (Amb; approximately 40 nL L21) and elevated (Elev; 100 nL L21) [O3] in each of the three growing seasons.Spectra were collected from July 14 to 17, 2013 (2013a), July 29 to 30, 2013 (2013b), July 14 to 18, 2014, and July 15 to 16, 2015.The number of individual rows sampled successfully during each time point is shown in parentheses. Asterisks indicate significanteffects of elevated [O3] within a given time point: **, P , 0.01; and ***, P , 0.001.








le Maire et al., 2004, 2008). Of course, this may be a dis-advantage as well, and further studies need to testwhether the PLSR models developed for maize in thisstudy also can be accurately used in other C4 species.

High-Throughput Phenotyping of the Stress Responsein Maize

This study used diverse maize lines grown at ele-vated [O3] as a proof-of-concept test bed for high-throughput phenotyping of leaf biochemical traitsusing leaf reflectance spectra. The maize inbred linesrepresent much of the genetic diversity within thespecies (McMullen et al., 2009), and elevated [O3], as atreatment, provides an attractive test for this approachenvironment because common phenotypes associatedwith O3 damage include reduced photosynthetic ca-pacity, leaf N content, and chlorophyll content, in-creased respiration rates and levels of antioxidants, andaltered SLA (Reich and Amundson, 1985; Fiscus et al.,2005; Betzelberger et al., 2012). Based on the biochem-ical and physiological traits estimated from hyper-spectral reflectance, both maize inbreds and hybridswere sensitive to elevated [O3] and showed reducedchlorophyll, N content, and photosynthetic capacity(Fig. 5; Table II). This was accompanied by greater SLA,indicating reduced leaf mass per unit area, which often

is associated with lower photosynthetic capacity and/or lower starch content (Poorter et al., 2012). This isconsistent with the accelerated leaf senescence at ele-vated [O3] observed in other species (Miller et al., 1999;Ainsworth et al., 2012; Gillespie et al., 2012) and withreduced carbon gain being at least partly responsiblefor the 10% reduction in yield of U.S. maize due to O3over the last 30 years (McGrath et al., 2015). This studyalso showed that, using hyperspectral reflectance, itwas possible to detect genotypic differences in physi-ological and biochemical leaf traits across maize linesand variation in their response to elevated [O3] (Fig. 5).The ability to rapidly detect this variation in the fieldacross a diverse panel of genotypes is a key first step toimproving abiotic stress tolerance in maize.

CONCLUSION

Rapid measurement of physiological traits is widelyanticipated as a transformative technology for the ex-perimental research needed to understand the mecha-nisms of crop stress response and to maximize theadvantage of genetic resources for crop improvement.Here, we provide evidence that leaf-level physiologicaland biochemical traits that are hallmark markers ofstress response can be accurately predicted from PLSRmodels built from leaf reflectance spectra in maize,

Table II. Statistical analysis (P values) of the effects of elevated [O3], genotype (Gen), and the interaction of [O3] and genotype on leaf physiologicaland biochemical traits estimated from leaf reflectance spectra

Box plots of leaf traits are shown in Figure 5. Spectra were collected from July 14 to 17, 2013 (2013a), July 29 to 30, 2013 (2013b), July 14 to 18,2014 (2014), and July 15 to 16, 2015 (2015). Inbred and hybrid lines were evaluated in different statistical models.

Parameter Chlorophyll N Vmax SLA Suc

Inbred lines2013a[O3] ,0.0001 0.17 0.56 ,0.0001 ,0.0001Gen ,0.0001 ,0.0001 ,0.0001 ,0.0001 ,0.0001[O3] 3 Gen 0.19 0.60 0.05 0.94 0.12

2013b[O3] ,0.0001 ,0.0001 0.95 ,0.0001 0.05Gen ,0.0001 ,0.0001 ,0.0001 ,0.0001 ,0.0001[O3] 3 Gen 0.90 0.35 0.72 0.98 0.99

2014[O3] ,0.0001 ,0.0001 ,0.0001 ,0.0001 0.004Gen ,0.0001 ,0.0001 ,0.0001 ,0.0001 ,0.0001[O3] 3 Gen 0.0002 0.05 0.03 ,0.0001 ,0.0001

2015[O3] ,0.0001 ,0.0001 0.0007 0.01 ,0.0001Gen ,0.0001 ,0.0001 ,0.0001 ,0.0001 ,0.0001[O3] 3 Gen 0.17 0.32 0.42 0.05 0.25

Hybrid lines2014[O3] ,0.0001 ,0.0001 ,0.0001 0.08 ,0.0001Gen 0.0005 0.12 0.0008 ,0.0001 ,0.0001[O3] 3 Gen 0.07 0.40 0.22 0.86 0.29

2015[O3] ,0.0001 0.68 ,0.0001 0.0007 ,0.0001Gen ,0.0001 ,0.0001 ,0.0001 0.004 0.02[O3] 3 Gen 0.74 0.90 0.33 0.74 0.75


Yendrek et al.



thereby providing a high-throughput, nondestructivemethod for phenotyping.

MATERIALS AND METHODS

Greenhouse Experiment

Maize (Zea mays) genotype B73 was planted on March 15, 2014, in 17-cm-diameter pots containing potting mix (Sunshine LC1 mix; Sun Gro Horticul-ture). Plants were grown in the greenhouse and fertilized with 40% LongAshton nutrient solution (Hewitt and Smith, 1975) with either 6 mM NH4NO3(ampleN) or 0.25mMNH4NO3 (lowN) starting 20 d after planting. The greenhousewas sunlit, and additional supplemental lighting from metal halide lamps pro-vided a minimum photosynthetic photon flux density of 200 mmol m22 s21 from6AM to 8 PM. Foliar tissue andhyperspectralmeasurementswere collected 55 to 60 dafter planting from a developmental range of leaves at V7 to V10.

Field Experiments

Diverse inbred and hybrid maize genotypes (Supplemental Table S4) weregrown in ambient and elevated [O3] at the Free Air Concentration Enrichmentfield site in Champaign, Illinois (soyface.illinois.edu) during the growing sea-sons of 2013, 2014, and 2015. Maize genotypes were planted with a precisionplanter in 3.35-m rows at a density of eight plants per meter and row spacing of0.76 m. In 2013, 203 inbred genotypes were grown at ambient and elevated [O3](n = 2) in octagonal rings approximately 20 m in diameter. Each individualgenotype was planted in a single row plot within a ring. Placement of indi-vidual genotypes was randomized between the two replicates, and the locationof each genotype was fixed within a pair of ambient and elevated [O3] rings(ring pair) that were located near one another in the field (but still separated byapproximately 100 m). B73 was planted in 12 locations within each ring in 2013.In 2014, 52 inbred and 26 hybrid genotypes were grown at ambient and ele-vated [O3] (n= 4). Each inbred genotypewas planted in a single rowwithin eachring, and the location of genotypes was randomized among the four replicatesbut conserved within an ambient and elevated [O3] ring pair. Each hybridgenotype was planted in two rows within each ring and randomized as de-scribed above for inbred genotypes. In 2014, B73 was planted in 10 rows withineach inbred ring and B73 3 Mo17 was planted in 14 rows within each hybridring. In 2015, 10 inbred and eight hybrid genotypes were grown at ambient andelevated [O3] (n = 4). Each genotype was planted in five rows within each oc-tagonal ring with a spatial design ensuring that each genotype was planted inone row in the central, north, south, east, and west sides of the ring.

O3was increased to a target set point of 100 nLL21 fromapproximately 10 AM to6 PM throughout the growing seasons using free-air O3 enrichment technology, asdescribed by Morgan et al. (2004) with the following modifications. O3 was pro-duced by passing high voltage through oxygen in O3 generators (CFS-3 2G; Ozo-nia) and then added to compressed air for fumigation.O3-enriched airwas releasedon the upwind sides of the octagon from horizontal pipes surrounding the rings.Within each ring, there was a 1-m buffer area that was not used for experiments.Fumigation was stopped when leaves were wet (during rain) and when windspeed was too low (less than 0.5 m s21) to maintain consistent treatment. The av-erage daily [O3] and cumulative [O3] exposures during the 2013, 2014, and2015 growing seasons are summarized in Supplemental Table S5.

Training Data Set

Training data sets were collected to pair leaf reflectance spectra with goldstandard biochemical and physiological measurements of chlorophyll content,N content, SLA, Vmax, Vp,max, Suc content, and ORAC. Measurements for thetraining data set were collected from diverse maize lines grown at ambient andelevated [O3] in the field in 2014 and 2015 and from inbred line B73 grown in thegreenhouse at high and lowN availability. The training data sets for each of thesuccessful PLSR traits are provided in Supplemental Data Set S1.

Gold Standard Methods for Measuring Leaf Physiologyand Biochemistry

To determineVp,max andVmax for the training data set, the response ofA toCiwas measured across a range of [CO2] from 25 to 1,500 mmol mol21 using aportable photosynthesis system (LI-6400; LICOR Biosciences) set at constant

photosynthetic photon flux density (2,000 mmol m22 s21). The approximatelysix initial points of the A/Ci curve (Ci , 60 mmol mol21) were fit to a model forphosphoenolpyruvate carboxylase-limited photosynthesis (von Caemmerer,2000) and used to estimated Vp,max. Vmax was estimated as the horizontal as-ymptote of a four-parameter nonrectangular hyperbolic function (Markelzet al., 2011). Before starting each A/Ci curve, leaf temperature was measuredwith an infrared thermometer (62MAX; Fluke) and used to set the leaf chambercuvette at ambient conditions. This also ensured that gas-exchange measure-ments were done at the same temperature as leaf reflectance spectra. For thegreenhouse experiment, two intact leaves were measured per plant: the firstand third youngest leaf with a visible collar. For field samples, the thirdyoungest leaf with a visible collar was excised and immediately recut underwater prior to dawn.

Immediately after eachA/Ci curvewas completed, tissuewas removed fromthe leaf using a cork borer and flash frozen in liquid N. One leaf disc (ap-proximately 1.4 cm2) was incubated in 96% (v/v) ethanol for 3 d at 4°C in orderto determine chlorophyll content using the equations of Lichtenthaler andWellburn (1983). Another leaf disc was used to determine the ORAC contentfollowing the procedures described by Gillespie et al. (2007). A third leaf discwas used to determine Suc content according to the methods of Jones et al.(1977). SLA was determined by weighing a known area of leaf tissue afterdrying for 5 d at 60°C. The dried leaf tissue was then ground to a fine powderand combusted with oxygen in an elemental analyzer (Costech 4010; CostechAnalytical Technologies). N content was determined by comparing experi-mental samples with an acetanalide standard curve.

Maize Leaf Reflectance Spectroscopy

Before destructively sampling leaf tissue, reflectance spectra were capturedfrom the adaxial surface of the same leaves that were used for gold standardmeasures of leaf physiology and biochemistry using a full-range spectroradi-ometer (ASD FieldSpec 4 Standard Res; Analytical Spectral Devices) equippedwith an illuminated leaf clip contact probe. The scanning time for the instrumentis approximately 100 ms, and each spectrum collected was the average of10 scans. A reflective white reference (Spectralon; Labsphere) was used tostandardize the relative reflectance of the samples. The spectroradiometer wasturned on for at least 30 to 60 min before reflectance measurements began toallow for the three arrays towarmup. Reflectancemeasurementswere obtainedfrom the same location on the leaf as the A/Ci curve measurements and wererestricted to the central section of the leaf, avoiding the midrib. A preliminarytest calculated bootstrapped (100-fold) mean and variance values from a pop-ulation of 25 measurements collected from a single leaf, from which we de-termined that six measurements per leaf reduced sample variance by over 50%(Supplemental Fig. S2). Subsequently, all leaf reflectance measurements usedthe average of six spectra from each leaf taken at different locations within anapproximately 7-cm band in the central section of the leaf. Before averaging thesix spectra collected from a single leaf, a splice correction was applied to thespectral observations to ensure continuous data across detectors. Data wereinterpolated to provide 1-nm bandwidths, and quality control was appliedusing the FieldSpectra package in R (Serbin et al., 2014).

PLSR Models

PLSR was used to develop predictive models with the open-source PLSpackage (Mevik andWehrens, 2007) in R following the methods of Serbin et al.(2014). This approach identifies latent factors, or components, that account formost of the variation in a trait variable and generates a linear model consistingof waveband scaling coefficients to transform full-spectrum data. To maximizethe predictive ability of the model while minimizing the possibility of over-fitting, the predicted residual sum of squares (PRESS) statistic was used todetermine the optimal number of model components (Chen et al., 2004). Thenumber of components that minimized the RMSE of the PRESS statistic waschosen (Wold et al., 2001). We calculated the PRESS statistic through a leave-one-out cross-validation approach, which trains the model on all but one ob-servation and then makes a prediction for the left-out observation. The averageerror is then computed and used to evaluate the model (Siegmann and Jarmer,2015). VIP scores (Wold, 1994) also were determined to compare the relative sig-nificance of each wavelength in its contribution to the final model. The spectrumrange for all modelswas 500 to 2,400 nm, exceptN content, whichwas restricted to1,500 to 2,400 nm because of well-known attributes of N absorption (Curran, 1989;Serbin et al., 2014). Performanceparameterswere generated to assess the predictiveability of eachmodel, including the coefficient of determination (r2) from the leave-one-out cross-validation method, RMSE, and model bias.





http://soyface.illinois.edu





A second evaluation of the PLSR model was performed with a holdout dataset. In this case, 80% of the data were used as a training data set for the esti-mation of leaf traits, and 20% of the data were used as a testing data set.

For chlorophyll, N content, SLA, Vmax, and Vp,max, training data sets werecollected from both field-grown and greenhouse-grownmaize plants. For thesetraits, two PLSR models were built, one with field-grown data only and a sec-ondwith the entire trainingdata set from thefield and the greenhouse (Fig. 1). Thebetter of the two PLSRmodels based on larger r2 and smaller RMSE (typically theone with data from both the field and greenhouse) was then applied to the O3screening data set collected in the field in 2013, 2014, and 2015. The PLSR modelcoefficients for each trait are provided in Supplemental Table S1.

High-Throughput Screening for O3 Response

As a test of the potential to use leaf reflectance spectra for high-throughputscreening, diverse maize germplasm were measured in the FACE experimentdescribed above in 2013, 2014, and 2015 (Supplemental Data Set S2). Leaf re-flectance spectra were taken on the third youngest leaf with a collar from allexperimental rows (1,024 in 2013, 768 in 2014, and 912 in 2015). In 2013 and2014, two plants per genotype (row) were sampled within each ambient andelevated [O3] ring, and in 2015, a single plant per row was sampled becauseeach genotype was planted in five rows within each ring. Leaf reflectance wasmeasured twice in 2013, from July 14 to 17 and July 29 to 30, and from July 14 to18, 2014, and July 15 to 16, 2015.

Statistical Analysis of Measured and PredictedTrait Values

In addition to PLSR models, published indices for estimating chlorophyllcontent, N content, and SLAwere tested using the training population (Table I).Correlations between published indices and measured values were tested.Pairwise correlations among the gold standard measured traits and thespectral-derived traits were tested using the training data set. The ability todistinguish statistical differences in measured and PLSR-modeled traits alsowas tested with a subset of the training data set. Measured and PLSR-modeledestimates of chlorophyll content, N content,Vmax, and SLA from genotypes B73,Mo17, and B73 3 Mo17 were used to test for the effects of genotype, O3, andtheir interaction using a general linear model.

The field designs in 2013, 2014, and 2015 differed in the number of gen-otypes examined and the layout of the FACE and control rings among years.Models were fit for each year separately. For the 2013 data, the linear modelYijkl = m + tj + rij + gk + tgij + «ijkl was fit, where j is the jth treatment (ambient,elevated [O3]), i is the random effect of the ith ring pair (i = 1,..4), and k is the kthgenotype. The variance of the ring pairs was assumed to be common for allambient rings and all elevated [O3] rings but not assumed to be the same be-tween them, and an additional covariance parameter for the southwest cornerof the ringswas included. In 2014,Yijkl =m+ tj + rij + gk + tgij + «ijkl was fit, where jis the jth treatment, i is the random effect of the ith ring pair (i = 1,..4), and k is thekth genotype. The increased number of replicates in 2014 allowed us to fit theresiduals for the difference in variance structure as « ; Ν(0,S), where S is thevariance covariancematrix with diagonal elements sj and all off-diagonal termsare zero except for a single covariance p to account for spatial variation amongobservations in the southwest quadrant of the elevated [O3] rings. In 2015,the ring was divided into five sectors, i.e. northeast, northwest, southeast,southwest and central, and in each sector all of the genotypes were planted. Themodel Yijkl = m + tj + rij + gs(pij) + bl + gk + tgij + «ijkl was fit, where Yijkl is the traitvalue for the ith ring pair (i = 1,..4), the sth sector (s = 1,.5) nested inside thering pair for the jth treatment, the lth sector (1 for the sector in the southwest,0 otherwise), and the kth genotype. Ring pair and sector are treated as randomwith all other effects fixed.

Supplemental Data

The following supplemental materials are available.

Supplemental Figure S1. Mean measured and PLSR-modeled leaf traits 61 SD for maize genotypes B73, B73 3 Mo17, and Mo17.

Supplemental Figure S2. Bootstrap estimate of mean chlorophyll contentand SD from one to 24 spectra taken on an individual leaf.

Supplemental Table S1. PLSR model coefficients for predicting foliar traitsusing leaf reflectance spectra from maize.

Supplemental Table S2. Performance comparison between predictivemodels built with either the entire training population (full model) or80% of the training population (80% model).

Supplemental Table S3. ANOVA degrees of freedom, F value, and signif-icance level for measured and PLSR-modeled leaf traits.

Supplemental Table S4. List of inbred and hybrid maize genotypes grownat ambient (approximately 40 nL L21) and elevated (approximately100 nL L21) [O3] in the 2013, 2014, and 2015 growing seasons.

Supplemental Table S5. Seasonal average daily [O3] exposure at theSoyFACE facility in 2013, 2014, and 2015.

Supplemental Data Set S1. Raw spectral data for wavelengths used tobuild PLSR models, along with measured trait values and PLSR-predictedtrait values.

Supplemental Data Set S2. Raw spectral data for the screening data setcollected on diverse maize inbred and hybrid lines grown in the field inambient and elevated [O3] in 2013, 2014, and 2015.

ACKNOWLEDGMENTS

We thank Brad Dalsing, Chad Lantz, Kannan Puthuval, and Mac Singer foroperating the SoyFACE facility and Matt Lyons, Dan Jaskowiak, John Regan,Anna Molineaux, Jessica Wedow, Taylor Pederson, Chris Moller, CharlesBurroughs, and Ben Thompson for assistance with field sampling and labora-tory measurements.

Received September 14, 2016; accepted November 10, 2016; published Novem-ber 15, 2016.

LITERATURE CITED

Adjorlolo C, Cho MA, Mutanga O, Ismail R (2012) Optimizing spectralresolutions for the classification of C3 and C4 grass species, usingwavelengths of known absorption features. J Appl Remote Sens 6:063560

Ainsworth EA, Beier C, Calfapietra C, Ceulemans R, Durand-Tardif M,Farquhar GD, Godbold DL, Hendrey GR, Hickler T, Kaduk J, et al(2008) Next generation of elevated [CO2] experiments with crops: a criticalinvestment for feeding the future world. Plant Cell Environ 31: 1317–1324

Ainsworth EA, Serbin SP, Skoneczka JA, Townsend PA (2014) Using leafoptical properties to detect ozone effects on foliar biochemistry. Photo-synth Res 119: 65–76

Ainsworth EA, Yendrek CR, Sitch S, Collins WJ, Emberson LD (2012) Theeffects of tropospheric ozone on net primary productivity and implica-tions for climate change. Annu Rev Plant Biol 63: 637–661

Araus JL, Cairns JE (2014) Field high-throughput phenotyping: the newcrop breeding frontier. Trends Plant Sci 19: 52–61

Asner GP, Martin RE (2008) Spectral and chemical analysis of tropicalforests: scaling from leaf to canopy levels. Remote Sens Environ 112:3958–3970

Asner GP, Martin RE (2015) Spectroscopic remote sensing of non-structural carbohydrates in forest canopies. Remote Sens 7: 3526–3547

Asner GP, Martin RE, Knapp DE, Tupayachi R, Anderson C, Carranza L,Martinez P, Houcheime M, Sinca F, Weiss P (2011) Spectroscopy ofcanopy chemicals in humid tropical forests. Remote Sens Environ 115:3587–3598

Atzberger C, Guerif M, Baret F, Werner W (2010) Comparative analysis ofthree chemometric techniques for the spectroradiometric assessment ofcanopy chlorophyll content in winter wheat. Comput Electron Agric 73:165–173

Betzelberger AM, Yendrek CR, Sun J, Leisner CP, Nelson RL, Ort DR,Ainsworth EA (2012) Ozone exposure response for U.S. soybean culti-vars: linear reductions in photosynthetic potential, biomass, and yield.Plant Physiol 160: 1827–1839

Brosché M, Merilo E, Mayer F, Pechter P, Puzõrjova I, Brader G, KangasjärviJ, Kollist H (2010) Natural variation in ozone sensitivity among Arabidopsisthaliana accessions and its relation to stomatal conductance. Plant Cell En-viron 33: 914–925


Yendrek et al.














Chen S, Hong X, Harris CJ, Sharkey PM (2004) Sparse modeling usingorthogonal forward regression with PRESS statistic and regularization.IEEE Trans Syst Man Cybern B Cybern 34: 898–911

Cheng T, Rivard B, Sanchez-Azofeifa AG, Feret JB, Jacquemoud S, UstinSL (2014) Deriving leaf mass per area (LMA) from foliar reflectanceacross a variety of plant species using continuous wavelet analysis.ISPRS J Photogramm Remote Sens 87: 28–38

Christensen JH, Hewitson B, Busuioc A, Chen A, Gao X, Held I, Jones R,Kolli RK, Kwon WT Laprise R, et al (2007) Regional climate projec-tions. In S Solomon, D Qin, M Manning, Z Chen, M Marquis, KB Averyt,M Tignor, HL Miller, eds, Climate Change 2007: The Physical ScienceBasis. Contribution of Working Group I to the Fourth Assessment Re-port of the Intergovernmental Panel on Climate Change. CambridgeUniversity Press, Cambridge, UK, pp 847–940

Curran PJ (1989) Remote sensing of foliar chemistry. Remote Sens Environ30: 271–278

Datt B (1998) Remote sensing of chlorophyll a, chlorophyll b, chlorophyll a+b and total carotenoid content in eucalyptus leaves. Remote Sens En-viron 66: 111–121

Dhondt S, Wuyts N, Inzé D (2013) Cell to whole-plant phenotyping: thebest is yet to come. Trends Plant Sci 18: 428–439

Dizengremel P, Le Thiec D, Bagard M, Jolivet Y (2008) Ozone risk as-sessment for plants: central role of metabolism-dependent changes inreducing power. Environ Pollut 156: 11–15

Dodd AN, Salathia N, Hall A, Kévei E, Tóth R, Nagy F, Hibberd JM,Millar AJ, Webb AAR (2005) Plant circadian clocks increase photo-synthesis, growth, survival, and competitive advantage. Science 309:630–633

Doughty CE, Asner GP, Martin RE (2011) Predicting tropical plant phys-iology from leaf and canopy spectroscopy. Oecologia 165: 289–299

Dreccer MF, Barnes LR, Meder R (2014) Quantitative dynamics of stemwater soluble carbohydrates in wheat can be monitored in the fieldusing hyperspectral reflectance. Field Crops Res 159: 70–80

Farquhar GD, von Caemmerer S, Berry JA (1980) A biochemical model ofphotosynthetic CO2 assimilation in leaves of C3 species. Planta 149: 78–90

Fiscus EL, Booker FL, Burkey KO (2005) Crop responses to ozone: uptake,modes of action, carbon assimilation and partitioning. Plant Cell Envi-ron 28: 997–1011

Frei M, Tanaka JP, Wissuwa M (2008) Genotypic variation in tolerance toelevated ozone in rice: dissection of distinct genetic factors linked totolerance mechanisms. J Exp Bot 59: 3741–3752

Furbank RT, Tester M (2011) Phenomics: technologies to relieve the phe-notyping bottleneck. Trends Plant Sci 16: 635–644

Gamon JA, Serrano L, Surfus JS (1997) The photochemical reflectance in-dex: an optical indicator of photosynthetic radiation use efficiencyacross species, functional types, and nutrient levels. Oecologia 112: 492–501

Gillespie KM, Chae JM, Ainsworth EA (2007) Rapid measurement of totalantioxidant capacity in plants. Nat Protoc 2: 867–870

Gillespie KM, Xu F, Richter KT, McGrath JM, Markelz RJC, Ort DR,Leakey ADB, Ainsworth EA (2012) Greater antioxidant and respiratorymetabolism in field-grown soybean exposed to elevated O3 under bothambient and elevated CO2. Plant Cell Environ 35: 169–184

Gitelson AA, Buschmann C, Lichtenthaler HK (1999) The chlorophyllfluorescence ratio F735/F700 as an accurate measure of the chlorophyllcontent in plants. Remote Sens Environ 69: 296–302

Gitelson AA, Merzlyak MN (1997) Remote estimation of chlorophyllcontent in higher plant leaves. Int J Remote Sens 18: 2691–2697

Großkinsky DK, Svensgaard J, Christensen S, Roitsch T (2015) Plantphenomics and the need for physiological phenotyping across scales tonarrow the genotype-to-phenotype knowledge gap. J Exp Bot 66: 5429–5440

Hansen PM, Schjoerring JK (2003) Reflectance measurement of canopybiomass and nitrogen status in wheat crops using normalized differencevegetation indices and partial least squares regression. Remote SensEnviron 86: 542–553

Herrmann I, Karnieli A, Bonfil DJ, Cohen Y, Alchanatis V (2010) SWIR-based spectral indices for assessing nitrogen content in potato fields. IntJ Remote Sens 31: 5127–5143

Hewitt EJ, Smith TA (1975) Plant Mineral Nutrition. English UniversitiesPress, London

Irisarri JGN, Oesterheld M, Verón SR, Paruelo JM (2009) Grass speciesdifferentiation through canopy hyperspectral reflectance. Int J RemoteSens 22: 5959–5975

Jacquemoud S, Baret F (1990) PROSPECT: a model of leaf optimal prop-erties spectra. Remote Sens Environ 34: 75–91

Jones MGK, Outlaw WH, Lowry OH (1977) Enzymic assay of 10 to10 moles of sucrose in plant tissues. Plant Physiol 60: 379–383

Kim KM, Kwon YS, Lee JJ, Eun MY, Sohn JK (2004) QTL mapping andmolecular marker analysis for the resistance of rice to ozone. Mol Cells17: 151–155

Kraft M, Weigel HJ, Mejer GJ, Brandes F (1996) Reflectance measurementsof leaves for detecting visible and non-visible damage to crops. J PlantPhysiol 148: 148–154

Leitao L, Bethenod O, Biolley JP (2007) The impact of ozone on juvenilemaize (Zea mays L.) plant photosynthesis: effects on vegetative biomass,pigmentation, and carboxylases (PEPc and Rubisco). Plant Biol (Stuttg)9: 478–488

le Maire G, Francois C, Dufrene E (2004) Towards universal broad leafchlorophyll indices using PROSPECT simulated database and hyper-spectral reflectance measurements. Remote Sens Environ 89: 1–28

le Maire G, Francois C, Soudani K, Berveiller D, Pontailler JY, Breda N,Genet H, Davi H, Dufrene E (2008) Calibration and validation of hy-perspectral indices for the estimation of broadleaved forest leaf chloro-phyll content, leaf mass per area, leaf area index and leaf canopybiomass. Remote Sens Environ 112: 3846–3864

Lichtenthaler HK, Wellburn AR (1983) Determination of total carotenoidsand chlorophylls a and b of leaf extracts in different solvents. BiochemSoc Trans 11: 591–592

Liu L, Cheng Z (2011) Mapping C3 and C4 plant functional types usingseparated solar-induced chlorophyll fluorescence from hyperspectraldata. Int J Remote Sens 32: 9171–9183

Maccioni A, Agati G, Mazzinghi P (2001) New vegetation indices for re-mote measurement of chlorophylls based on leaf directional reflectancespectra. J Photochem Photobiol B 61: 52–61

Markelz RJC, Strellner RS, Leakey ADB (2011) Impairment of C4 photo-synthesis by drought is exacerbated by limiting nitrogen and amelio-rated by elevated [CO2] in maize. J Exp Bot 62: 3235–3246

Martin ME, Aber JD (1997) High spectral resolution remote sensing offorest canopy lignin, nitrogen and ecosystem processes. Ecol Appl 7:431–443

McGrath JM, Betzelberger AM, Wang S, Shook E, Zhu XG, Long SP,Ainsworth EA (2015) An analysis of ozone damage to historical maizeand soybean yields in the United States. Proc Natl Acad Sci USA 112:14390–14395

McMullen MD, Kresovich S, Villeda HS, Bradbury P, Li H, Sun Q, Flint-Garcia S, Thornsberry J, Acharya C, Bottoms C, et al (2009) Geneticproperties of the maize nested association mapping population. Science325: 737–740

McMurtey JE III, Chapelle EW, Kim MS, Meisinger JJ, Corp LA (1994)Distinguishing nitrogen fertilization levels in field corn (Zea mays L.)with actively induced fluorescence and passive reflectance measure-ments. Remote Sens Environ 47: 36–44

Meroni M, Panigada C, Rossini M, Picchi V, Cogliati S, Colombo R (2009)Using optical remote sensing techniques to track the development ofozone-induced stress. Environ Pollut 157: 1413–1420

Mevik BH, Wehrens R (2007) The pls package: principal component andpartial least squares regression in R. J Stat Softw 18: 1–24

Miller JD, Arteca RN, Pell EJ (1999) Senescence-associated gene expressionduring ozone-induced leaf senescence in Arabidopsis. Plant Physiol 120:1015–1024

Morgan PB, Bernacchi CJ, Ort DR, Long SP (2004) An in vivo analysis ofthe effect of season-long open-air elevation of ozone to anticipated2050 levels on photosynthesis in soybean. Plant Physiol 135: 2348–2357

Penuelas J, Baret F, Filella I (1995) Semi-empirical indices to assess ca-rotenoids/chlorophyll a ratio from leaf spectral reflectance. Photo-synthetica 31: 221–230

Penuelas J, Filella I (1998) Visible and near-infrared reflectance techniquesfor diagnosing plant physiological status. Trends Plant Sci 3: 151–156

Poorter H, Niklas KJ, Reich PB, Oleksyn J, Poot P, Mommer L (2012)Biomass allocation to leaves, stems and roots: meta-analyses of inter-specific variation and environmental control. New Phytol 193: 30–50

Reich PB, Amundson RG (1985) Ambient levels of ozone reduce netphotosynthesis in tree and crop species. Science 230: 566–570





Reich PB, Walters MB, Ellsworth DS (1997) From tropics to tundra: globalconvergence in plant functioning. Proc Natl Acad Sci USA 94: 13730–13734

Richardson AD, Reeves JB (2005) Quantitative reflectance spectroscopy as analternative to traditional wet lab analysis of foliar chemistry: near-infrared andmid-infrared calibrations compared. Can J Res 35: 1122–1130

Rudorff BFT, Mulchi CL, Daughtry CST, Lee EH (1996) Growth, radiationuse efficiency, and canopy reflectance of wheat and corn grown underelevated ozone and carbon dioxide atmospheres. Remote Sens Environ55: 163–173

Runeckles VC, Resh HM (1975) The assessment of chronic ozone injury toleaves by reflectance spectrophotometry. Atmos Environ 9: 447–452

Serbin SP, Dillaway DN, Kruger EL, Townsend PA (2012) Leaf opticalproperties reflect variation in photosynthetic metabolism and its sensi-tivity to temperature. J Exp Bot 63: 489–502

Serbin SP, Singh A, Desai AR, Dubois SG, Jablonski AD, Kingdon CC,Kruger EL, Townsend PA (2015) Remotely estimating photosyntheticcapacity, and its response to temperature, in vegetation canopies usingimaging spectroscopy. Remote Sens Environ 167: 78–87

Serbin SP, Singh A, McNeil BE, Kingdon CC, Townsend PA (2014)Spectroscopic determination of leaf morphological and biochemicaltraits for northern temperate and boreal tree species. Ecol Appl 24: 1651–1669

Serrano L, Penuelas J, Ustin SL (2002) Remote sensing of nitrogen andlignin in Mediterranean vegetation from AVIRIS data: decomposingbiochemical from structural signals. Remote Sens Environ 81: 355–364

Siegmann B, Jarmer T (2015) Comparison of different regression modelsand validation techniques for the assessment of wheat leaf area indexfrom hyperspectral data. Int J Remote Sens 36: 4519–4534

Sims DA, Gamon JA (2002) Relationships between leaf pigment contentand spectral reflectance across a wide range of species, leaf structuresand developmental stages. Remote Sens Environ 81: 337–354

Street NR, James TM, James T, Mikael B, Jaakko K, Mark B, Taylor G(2011) The physiological, transcriptional and genetic responses of anozone-sensitive and an ozone tolerant poplar and selected extremes oftheir F2 progeny. Environ Pollut 159: 45–54

Tilman D, Balzer C, Hill J, Befort BL (2011) Global food demand and thesustainable intensification of agriculture. Proc Natl Acad Sci USA 108:20260–20264

Ueda Y, Siddique S, Frei M (2015) A novel gene, OZONE-RESPONSIVEAPOPLASTIC PROTEIN1, enhances cell death in ozone stress in rice.Plant Physiol 169: 873–889

Vogelmann JE, Rock BN, Moss DM (1993) Red edge spectral measure-ments from sugar maple leaves. Int J Remote Sens 14: 1563–1575

von Caemmerer S (2000) Biochemical Models of Leaf Photosynthesis.CSIRO Publishing, Collingwood, Australia

Weber VS, Araus JL, Cairns JE, Sanchez C, Melchinger AE, Orsini E (2012)Prediction of grain yield using reflectance spectra of canopy and leaves inmaize plants grown under different water regimes. Field Crops Res 128: 82–90

Wold S (1994) PLS for multivariate linear modeling. In H van de Water-beernd, ed, Chemometric Methods in Molecular Design, Methods andPrinciples in Medicinal Chemistry. Verlag-Chemie, Weinheim, Ger-many, pp 195–218

Wold S, Sjostrom M, Eriksson L (2001) PLS-regression: a basic tool ofchemometrics. Chemometr Intell Lab 58: 109–130

Wright IJ, Reich PB, Westoby M, Ackerly DD, Baruch Z, Bongers F,Cavender-Bares J, Chapin T, Cornelissen JHC, Diemer M, et al (2004)The worldwide leaf economics spectrum. Nature 428: 821–827


Yendrek et al.



high-throughput phenotyping of maize leaf physiological ... · high-throughput, noninvasive ﬁeld...

Documents